Abstract This PhD thesis explores recent advancements in quantitative radiology applied to the diagnosis and characterization of myopathies using magnetic resonance imaging (MRI). The work integrates artificial intelligence (AI), radiomics, and advanced imaging techniques to improve diagnostic accuracy and pattern recognition in muscle diseases. The thesis is structured into three main studies. First, a systematic review evaluates the state of AI-assisted pattern recognition in MRI of myopathies, highlighting the potential of deep learning and machine learning models to outperform human experts in distinguishing between myopathy subtypes, despite heterogeneity in datasets and methodologies. Second, an original radiomic study applies texture analysis to whole-thigh fat fraction (FF) maps derived from Dixon sequences in a cohort of patients with genetically or histologically confirmed myopathies. Manual segmentation and PyRadiomics feature extraction enabled unsupervised clustering, revealing two distinct radiomic patterns corresponding to different clinical phenotypes. The study demonstrates that volumetric radiomics can identify pathophysiological differences beyond visual interpretation, especially when applied to full muscle volumes rather than single slices. Third, the thesis presents a multidisciplinary study of the split-hand syndrome (SHS) in Charcot-Marie-Tooth disease type X1 (CMTX1), combining clinical, neurophysiological, and radiological data. SHS, previously associated mainly with motor neuronopathies, was identified in over half of CMTX1 patients using standard criteria and FF-based MRI metrics, suggesting a broader pathological spectrum and underlining the importance of imaging in early and differential diagnosis. Altogether, the thesis underscores the transformative potential of AI and radiomics in neuromuscular imaging, proposing reproducible, quantitative, and non-invasive tools for diagnosis and disease monitoring. Future directions include multicenter collaborations, automated segmentation, and integration with explainable AI to enhance clinical applicability.

NEW ADVANCES IN QUANTITATIVE RADIOLOGY:MRI IMAGING IN MYOPATHIES

MOSCATELLI, MARCO ELVIO MANLIO
2025

Abstract

Abstract This PhD thesis explores recent advancements in quantitative radiology applied to the diagnosis and characterization of myopathies using magnetic resonance imaging (MRI). The work integrates artificial intelligence (AI), radiomics, and advanced imaging techniques to improve diagnostic accuracy and pattern recognition in muscle diseases. The thesis is structured into three main studies. First, a systematic review evaluates the state of AI-assisted pattern recognition in MRI of myopathies, highlighting the potential of deep learning and machine learning models to outperform human experts in distinguishing between myopathy subtypes, despite heterogeneity in datasets and methodologies. Second, an original radiomic study applies texture analysis to whole-thigh fat fraction (FF) maps derived from Dixon sequences in a cohort of patients with genetically or histologically confirmed myopathies. Manual segmentation and PyRadiomics feature extraction enabled unsupervised clustering, revealing two distinct radiomic patterns corresponding to different clinical phenotypes. The study demonstrates that volumetric radiomics can identify pathophysiological differences beyond visual interpretation, especially when applied to full muscle volumes rather than single slices. Third, the thesis presents a multidisciplinary study of the split-hand syndrome (SHS) in Charcot-Marie-Tooth disease type X1 (CMTX1), combining clinical, neurophysiological, and radiological data. SHS, previously associated mainly with motor neuronopathies, was identified in over half of CMTX1 patients using standard criteria and FF-based MRI metrics, suggesting a broader pathological spectrum and underlining the importance of imaging in early and differential diagnosis. Altogether, the thesis underscores the transformative potential of AI and radiomics in neuromuscular imaging, proposing reproducible, quantitative, and non-invasive tools for diagnosis and disease monitoring. Future directions include multicenter collaborations, automated segmentation, and integration with explainable AI to enhance clinical applicability.
15-lug-2025
Inglese
SCONFIENZA, LUCA MARIA
DEL FABBRO, MASSIMO
Università degli Studi di Milano
42
File in questo prodotto:
File Dimensione Formato  
phd_unimi_R13443.pdf

accesso aperto

Dimensione 1.76 MB
Formato Adobe PDF
1.76 MB Adobe PDF Visualizza/Apri

I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.14242/217793
Il codice NBN di questa tesi è URN:NBN:IT:UNIMI-217793